Researchers typically use moderated regression models to examine the presence of linear moderation effects in their studies. However, researchers rarely conduct a robustness check following a significant moderation effect to investigate whether the moderation effect is spurious. The misleading moderation can occur when a predictor and a moderator variable correlate and the true nature of the relationships between predictors and a dependent variable are nonlinear. In this paper, we propose and illustrate the use of an information theoretic approach in moderation analysis with the aim of avoiding spurious moderation effects. We demonstrate our suggested procedure using Monte Carlo simulations and real data from published studies.